Salesforce's job-ready agents and Winter '27: what to do before Agentforce switches itself on
Salesforce has launched a portfolio of role-specific Agentforce agents, while Winter '27 quietly enables Agentforce in eligible orgs. Here is what is real, what is vendor-reported, and the admin checks worth doing now.
Two Salesforce developments are landing at once, and they are easy to treat as separate stories. One is a new set of ready-made, role-specific Agentforce agents. The other is the Winter '27 release, which is making Agentforce tooling available in eligible orgs as part of the upgrade. Together they change the question for many Salesforce customers from "should we try agents?" to "who is allowed to build one, and what should it do first?"
What Salesforce announced
In its 14 September release, Salesforce introduced a portfolio of agents designed around specific jobs rather than a blank canvas:
- Casey handles customer service across voice, SMS, WhatsApp and web chat.
- Paige resolves employee IT and HR requests.
- Carter helps shoppers find, compare and check out.
- Marshall runs back-office supply chain processes, with deterministic execution and an audit record of every action.
- Piper engages and qualifies inbound leads.
- Fin handles complex customer experience workflows across channels.
- Hunter works an outbound sales pipeline over weeks and months.
Salesforce lists all but Hunter as generally available now. Hunter is in pilot, with general availability targeted for November. Hunter is also the first agent on a new long-horizon runtime, built on memory across sessions, durable execution that can be resumed or corrected, and steering based on feedback. Salesforce says more agents will follow, but gives no date. Multi-Agent Orchestration, which routes work between specialised agents, is listed as generally available, and AI Skills and Agent Optimizer are targeted for October.
Winter '27: enablement is not activation
According to CET Digit's guide to the release, Agentforce is being enabled automatically in eligible editions at no extra cost, with production upgrades on the weekends of 3 and 10 October. Enablement makes the Agentforce Builder and related tools available to users with the right permissions. It does not switch any agent on or, by itself, create consumption charges. Agent actions do consume credits once agents run, so the cost arrives when someone builds and activates something.
That distinction matters. The risk is not that an agent starts talking to your customers uninvited. It is that capability appears in your org before anyone has decided who owns it, how it is governed or what it should cost.
How much to believe the results
The headline numbers in Salesforce's release are striking: 90% of core shopper journeys handled by one retailer's agent, 70% of administrative requests resolved at Autism Queensland, 79% of conversations resolved autonomously at Anthropic. Futurum's analyst calls them production-grade rather than pilot results. They are also chosen by the vendor, from a small number of named customers, and not independently verified. Futurum itself notes that how widely they replicate is still open. Treat them as proof that the approach can work well in the right workflow, not as a forecast for yours.
Futurum also points to a strategic effect. The more agents a customer deploys, the more the orchestration and optimisation layers add value, and the higher the cost of switching. That is not a reason to avoid the platform, but it is a reason to keep your agent logic, data and evaluation material portable where you can.
Practical steps
- Confirm your upgrade date on Salesforce Trust, and name one owner for release readiness.
- Audit who holds the Manage AI Agents permission. Limit it to the people who should be building agents.
- Check your Einstein settings are configured as you intend, since the master setting affects the whole platform.
- Test the enforced release updates in a preview sandbox, including SOAP login integrations and the profile visibility change.
- Pick one governed first use case with a named business owner, a clear handoff to a person and a measure of success you define yourself.
- Treat agent credits like cloud spend. Set a budget, assign an owner and review consumption monthly.
- Test with your own data and edge cases, not the vendor's case studies.
Pre-built agents genuinely lower the barrier, and for common workflows like service, inbound leads and employee requests they are worth a serious look. They do not remove the need for clean data, sensible permissions and someone accountable. An independent adviser can help scope a first use case, review the permission and security model, and assess the real cost before a pilot becomes a commitment.
The takeaway: the platform is making it easy to start. The organisations that benefit will be the ones that decide deliberately what to start with.
Sources: Salesforce, "Salesforce Expands Agentforce With a New Portfolio of AI Agents Built for High-Value Work"; CET Digit, "Agentforce Turns Itself On: What Salesforce's Winter '27 Release Means for Your Org Before the October Go-Live"; Futurum, "Salesforce's Job-Ready Agents Target Enterprise AI's Biggest Gap".